Mirage (AgentVFS)

# Fix list: Mirage (AgentVFS)

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/agentvfs, the October 2026 research run, assessed 21 October 2026. Grade B, 65.3 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Mirage (AgentVFS): work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 57 out of 100, up to 7.5 more on the total

Why it scored 57: Read with the local-software and framework lines. Credentials belong to each mount and resolve from the environment, dotenv or AWS Secrets Manager, with 1Password and Auth0 as custom sources, so each virtual CLI can carry its own account. The server's `jwt` mode checks expiry, issuer and audience and rotates through JWKS, but the default `local` mode accepts unauthenticated requests on localhost when no token is set (18). Mount modes, `allow`, `ask` and `deny` command rules, `hide` and `show` path rules and pre-command policy scripts, with ask as an approval step (18). It returns untrusted content from mail, chat and documents, and no prompt-injection guidance was found in the docs; hidden paths and refusals limit what an injected command can reach (5). The source keeps a per-session log of operations and commands with timestamps in a disk or Redis store, little described in the docs (9). SECURITY.md with a disclosure policy and stated response times; no security.txt on strukto.ai (404), no bug bounty or certification, and no advisories on OSV for the packages (7).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Reliability, 63 out of 100, up to 7.4 more on the total

Why it scored 63: Read with the local-software lines, because Mirage runs in the owner's process or on the owner's server. Official packages `mirage-ai` on PyPI and `@struktoai/mirage-node`, `-browser`, `-agents` and `-cli` on npm, all 0.0.7, with Python 3.11 or later and Node.js engines stated (20). Eleven GitHub workflows, and the Python, TypeScript, Rust and integration test workflows showed passing on main on 21 October 2026 (25). 41 open issues and one open pull request against 4,741 commits in the last 90 days. We counted them and didn't read them, so crash and regression reports couldn't be told apart (15). Seven tags from 0.0.1 to 0.0.7 with one-line messages and no changelog, so breaking changes aren't called out (3). Labelled preview in the README and Development Status 3 - Alpha in pyproject.toml (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 3. Schema & documentation, 68 out of 100, up to 5.2 more on the total

Why it scored 68: Read as a library with an MCP server. Typed Python (py.typed) and TypeScript packages, and the seven MCP tools have typed inputs listed in the HTTP docs. No OpenAPI file for the HTTP API (18). llms.txt at docs.mirage.strukto.ai indexes 284 Markdown pages, and a docs MCP server searches them (10). The agents page says which connection suits which client, and each backend has a setup page with its credentials and limits (15). Mount modes are an enum (read, write, exec), profile rules are typed lists, and virtual CLI subcommands take JSON arguments whose schema `--help` prints (11). Quickstarts in three forms, an examples directory and a troubleshooting page; HTTP statuses such as 409 and refusal metadata are documented, without an error catalogue (10). Versioned packages and git tags, no changelog (4).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 4. Transparency & trust, 42 out of 100, up to 5.1 more on the total

Made of editorial 59, provenance 24.

Why it scored 42: Apache-2.0, an OSI licence, in LICENSE and the npm metadata. The VFS-Bench harness is MIT. The VFS.md format on agentvfs.org carries no licence, but it isn't the graded surface (30). Nothing is hosted by Strukto.AI. Data goes to the services the owner mounts and snapshots store secrets redacted, per the HTTP docs. strukto.ai's privacy policy covers the website (PostHog and Vercel analytics), and no statement says what the software sends (15). No deprecation policy; SECURITY.md supports 0.0.x only and the news page marks old posts outdated without dates (2). No telemetry found in the source or the docs, and no statement either way (12).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Legal entity named: not found (0 of 20)
- Domain age: strukto.ai, no registry record we could read (0 of 15)
- Status page: not found (0 of 10)
- Changelog: not found (0 of 10)
- security.txt: not found (0 of 10)

## 5. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: Read with the self-hosted rule. Free Apache-2.0 software with nothing to buy from Strukto.AI and no payment protocol (0). No price to publish (20). No card, account or trial (20). An agent can install the package and run it with no sign-up (20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 6. Agent ergonomics, 79 out of 100, up to 3.4 more on the total

Why it scored 79: Seven MCP tools (`shell`, `read`, `write`, `edit`, `ls`, `grep`, `glob`), and a session is offered only the ones its profile allows. In the library route the agent gets one shell tool, and AgentVFS adds a VFS.md that describes the mounts (23). `read` takes a line range, shell output streams in chunks of at most 8 KiB with a byte queue, and pipes, `head` and `jq` size output before it reaches the model (16). Refusals return metadata and an explain page says why a rule fired; exit codes are returned. No catalogue of error codes (14). `read`, `ls`, `grep` and `glob` carry `readOnlyHint`. Asks record a pending decision and the line is retried in the same session; snapshots can be restored. No destructive hint or idempotency key was found for writes (12). A RAM mount needs no credentials; Python and TypeScript SDKs plus a CLI (14).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 7. Maintenance & community, 88 out of 100, up to 1.1 more on the total

Why it scored 88: Version 0.0.7 tagged on 20 October 2026, the day before the check (30). 0.0.5 on 14 August, 0.0.6 on 3 September and 0.0.7 on 20 October, three in 90 days (20). Pull requests from nine accounts outside the strukto-ai organisation were merged in the last 90 days and one pull request is open. Issue reply times weren't read, because GitHub's web pages and API weren't reachable from this session (16). Strukto.AI's own Python and TypeScript packages are current at the same version (13). uv and pnpm lockfiles, Dependabot commits and passing CI; no PyPI release history was readable (9).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- AgentVFS itself is an interface and a Markdown format with a benchmark, not a package. This listing grades Mirage, its authors' implementation. If another implementation appears, AgentVFS may need its own entry or a protocol reading the site doesn't have outside payments.
- VFS-Bench is the authors' own benchmark of their own system, run with one trial per arm, on 200 tasks they selected from MCP-Atlas and Toolathlon with a VFS.md and SKILL.md they wrote per task. It hasn't been rerun by anyone else that we found, and it isn't peer reviewed (the citation points to the blog post).
- unchecked: the open issues' content and reply times, because GitHub's web pages and API weren't reachable from this session and issue text isn't read.
- unchecked: the PyPI release history, because pypi.org answered with a JavaScript check; the version 0.0.7 and weekly downloads come from shields.io badges.
- unchecked: GitHub release notes for the tags.
- Strukto.AI's "world's first virtual terminal for AI agents" is the vendor's claim and wasn't tested.

## Weaknesses

- Version 0.0.7, labelled preview and alpha, with no changelog or deprecation policy
- The server accepts unauthenticated requests on localhost in its default mode when no token is set
- No prompt-injection guidance for the mail, chat and document content it returns
- Its published benchmark is the authors' own, run once per arm on tasks they selected

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Read the workspace's VFS.md first and point grep, find and cat at the path it names, not at a whole mount.
- Run a virtual CLI's subcommand with `--help` to get its argument schema before calling it; subcommands take JSON arguments.
- Expect a refusal with metadata when a profile asks or denies; the decision must be answered and the line retried in the same session.
- Set `MIRAGE_AUTH_MODE=token` or `jwt` before exposing the server beyond localhost.

## What the review panel asked for

- a dated changelog per release
- a written deprecation policy
- mark write tools destructive
- publish a changelog
- an independent VFS-Bench rerun
- a changelog per release
- token auth by default

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.